CentOS 8 Docker环境中systemctl命令失效问题求助

问题背景 #

在Docker容器中安装Nuance Speech Suite时，安装脚本调用systemctl命令启动服务，出现错误：

System has not been booted with systemd as init system (PID 1). Can't operate. Failed to connect to bus: Host is down

使用以下命令构建镜像：

docker build -t myapp:11 .

即使添加--privileged或--cap-add=NET_ADMIN参数启动容器，执行systemctl仍报错：

docker run -it --privileged app:11 /bin/bash

docker run -it --cap-add=NET_ADMIN nuance:11 /bin/bash

Dockerfile内容 #

# syntax=docker/dockerfile:1
FROM centos:latest
USER root

RUN cd /etc/yum.repos.d/
RUN sed -i 's/mirrorlist/#mirrorlist/g' /etc/yum.repos.d/CentOS-*
RUN sed -i 's|#baseurl=http://mirror.centos.org|baseurl=http://vault.centos.org|g' /etc/yum.repos.d/CentOS-*

RUN yum -y update && \
    yum clean all
    
RUN yum -y install \
    java-11-openjdk-devel \
    perl-Data-Dumper \
    redhat-lsb-core.x86_64 \
    glibc.x86_64 \
    glibc.i686 \
    libstdc++.x86_64 \
    libstdc++.i686 \
    openssl \
    libgcc \
    libgcc.i686 \
    libaio.x86_64 \
    libaio.i686 \
    libnsl.i686 \
    ncurses-libs \
    httpd.x86_64 \
    unzip \
    -x postfix \
    -x mariadb-libs \
    zlib.i686 \
    zlib.x86_64

WORKDIR /tmp

COPY Nuance_Speech_Suite-11.0.10-x86_64-linux.tgz ./Nuance_Speech_Suite-11.0.10-x86_64-linux.tgz
COPY NRec-fr-FR-10.0.0-10.1.0.i686-linux.tar.gz ./languages/NRec-fr-FR-10.0.0-10.1.0.i686-linux.tar.gz
COPY NVE_fr_FR_audrey-ml_xpremium-2.1.0_linux.zip ./languages/NVE_fr_FR_audrey-ml_xpremium-2.1.0_linux.zip
COPY NRec-fr-FR-10.0.0-10.1.0-CumulativePatch-1_linux.zip ./languages/NRec-fr-FR-10.0.0-10.1.0-CumulativePatch-1_linux.zip
COPY NRec-fr-FR-10.0.0-10.1.0-CumulativePatch-2_linux.zip ./languages/NRec-fr-FR-10.0.0-10.1.0-CumulativePatch-2_linux.zip
COPY nuance.lic ./nuance.lic

RUN tar -zxf Nuance_Speech_Suite-11.0.10-x86_64-linux.tgz
RUN tar -zxf languages/NRec-fr-FR-10.0.0-10.1.0.i686-linux.tar.gz
RUN unzip languages/NVE_fr_FR_audrey-ml_xpremium-2.1.0_linux.zip
RUN unzip languages/NRec-fr-FR-10.0.0-10.1.0-CumulativePatch-1_linux.zip
RUN unzip languages/NRec-fr-FR-10.0.0-10.1.0-CumulativePatch-2_linux.zip

WORKDIR /tmp/Nuance_Speech_Suite-11.0.10

RUN ./setup.sh -s -f "/tmp/nuance.lic" -j "/usr/lib/jvm/java-11-openjdk" -V "/tmp/languages" -I "NLM,NSS"

构建日志错误片段 #

2022-12-16 09:22:11 setup.sh: info: Restarting the Nuance License Manager service
2022-12-16 09:22:11 setup.sh: info: starting command 'systemctl restart nuance-licmgr'; output sent to log
System has not been booted with systemd as init system (PID 1). Can't operate.
Failed to connect to bus: Host is down
2022-12-16 09:22:11 setup.sh: info: Command 'systemctl restart nuance-licmgr' returned 1
2022-12-16 09:22:11 setup.sh: error: install_postprocessing_nlm_startservices() failed to start services
2022-12-16 09:22:11 setup.sh: info: skipping invocation of install_postprocessing_nms() due to previous post processing errors
2022-12-16 09:22:11 setup.sh: info: Skipping install_execute_installsuite due to previous errors

解决办法 #

方法1：让容器以systemd为PID 1运行 #

Docker容器默认PID 1不是systemd，需修改启动方式让systemd成为init进程：

1. 修改Dockerfile，在末尾添加：

   CMD ["/usr/sbin/init"]
   

2. 构建镜像后，用以下命令启动容器：

   docker run -d --privileged -v /sys/fs/cgroup:/sys/fs/cgroup:ro myapp:11
   

   --privileged和挂载/sys/fs/cgroup是systemd正常运行的必要条件。

方法2：修改安装脚本跳过systemctl调用 #

若不想用systemd管理容器服务，可修改setup.sh脚本：

1. 解压安装包后，编辑setup.sh，找到调用systemctl restart nuance-licmgr的部分，注释掉或替换为直接启动服务的命令（如/path/to/nuance-licmgr start）。
2. 查看./setup.sh --help是否有跳过服务启动的参数（如--no-start-services），若有则添加该参数运行脚本。

方法3：用supervisord管理服务 #

放弃systemd，改用supervisord管理Nuance服务：

1. 在Dockerfile中安装supervisord：

   RUN yum install -y supervisor && mkdir -p /var/log/supervisor
   

2. 创建/etc/supervisord.conf配置文件，添加Nuance服务启动配置：

   [supervisord]
   nodaemon=true
   
   [program:nuance-licmgr]
   command=/path/to/nuance-licmgr start
   autorestart=true
   stdout_logfile=/var/log/supervisor/nuance-licmgr.log
   stderr_logfile=/var/log/supervisor/nuance-licmgr.err.log
   

3. 修改Dockerfile的CMD为启动supervisord：

   CMD ["/usr/bin/supervisord", "-c", "/etc/supervisord.conf"]
   

内容的提问来源于stack exchange，提问作者jeyGey